Abstract:

In order to solve the real-time problem of visual navigation system with traditional motion estimation algorithm, a new approach based on classified feature points for mobile robot motion estimation was proposed. For dividing feature points into far points and near points, the distances between feature points and mobile robot were calculated according to the 3-dimensional coordinates of feature points. The far points were sensitive to the rotational movement of robot, thus they were used to calculate rotational matrix; the near points were sensitive to translational motion, thus they were used to calculate the translational matrix. When the far points and the near points are 30% of original feature points, the proposed approach had equivalent accuracy but reduced 60% computing time compared with RANdom SAmple Consensus (RANSAC). The results demonstrate that, by using classified feature points, the proposed algorithm can effectively reduce computing time, meanwhile ensure accuracy of motion estimation, and it can meet the the real-time requirement with large feature points.

Key words: mobile robot, visual navigation, feature detection, motion estimation, RANdom SAmple Consensus (RANSAC) algorithm

摘要：

为解决移动机器人视觉导航系统在进行机器人运动估计时使用传统运动估计算法计算时间较长而导致实时性较差的问题,提出了一种基于特征点分类策略的移动机器人运动估计方法。根据移动机器人视觉导航系统提供的特征点三维坐标计算出特征点与机器人的距离,从而将特征点分为远点与近点。远点对于机器人的旋转运动是敏感的,因此可用于计算移动机器人的旋转矩阵;近点对于机器人的平移运动是敏感的,因此可用于计算机器人的平移矩阵。仿真实验中,当远点与近点数为原特征点数目的30%时,基于特征点分类策略的运动估计计算精度与传统RANSAC算法相当,并能减少60%的计算时间。仿真结果表明,基于特征点分类策略的运动估计方法能在不降低计算精度的前提下有效减少计算时间,在特征点数目较多时也能很好地适应实时性要求。

关键词: 移动机器人, 视觉导航, 特征检测, 运动估计, 随机采样一致算法